IXN 1 — Exploratory Data Analysis

#IMPORT (wrangled) data 
raw_data <- read_csv("data/wrangled_utterances_representations.csv") 

#WRANGLE into DF of utterance-insights [eg not unique utterances, 1 obs for each utterance_detail-code]
df_insights <- raw_data %>% 
  #rename and factorize columns
  mutate(
    sid = factor(UID), #NOT actually a unique utterance id, treat as sheet order id
    pid = factor(PID, levels = c( #define level order so happiness first
          "bjs827ee1u", "3r2sh20ei", "4728sjuiz", "7ACC0B75","92ghd48xe","iurmer28", "s294hoei", #HAPPINESS-FIRST
          "j2719eertu2","lkin27js09b","li832lin23","7382kwtue","E1D39056","8v892iige")),   #SPACE-FIRST
    
    utterance = Utterance,
    uid = factor(as.numeric(factor(paste(pid,factor(utterance))))), #construct a unique ID for utterances
    
    TASK = factor(recode(Condition, "Static"="static", "Interactive"="ixn" )),
    TASK = factor(TASK, levels = c("static", "ixn")), #reorder factor levels
    DATASET = factor(recode(Notebook, "Happiness"="happiness", "Space"="space")), #cleanup diff case
    outcomeType = recode(DATASET, "happiness"="numeric", "space"="nominal"),
    data_order = factor(paste(TASK,"_",DATASET)), #create an order var 
    data_order = recode(data_order, "ixn _ happiness"="space-first",
                                  "ixn _ space"="happiness-first",
                                  "NA _ NA"="NA",
                                  "static _ happiness"="happiness-first",
                                  "static _ space"="space-first"),
    top_code = factor(highlevel),
    #recode process 
    top_code = recode(top_code, "ANALYSIS PROCESS" = "PROCESS"),
    top_code = factor(top_code, levels = c("PROCESS","DATASET","VARIABLE","RELATIONSHIP")),
    mid_code = factor(`Data Type`),
    low_code = factor(UtteranceType),
    timestamp = Timestamp,
    repns = group,
    ixn = factor(interaction_used) #was interaction used?
  ) %>% select( #select only needed columns
    sid,uid,pid, TASK, DATASET, data_order, ixn, top_code, mid_code, low_code, repns, timestamp, utterance
  )

#DF OF UNIQUE UTTERANCES
df_uniques <- df_insights %>% select(uid, pid, TASK, DATASET) %>% 
  distinct()  #take only unique utterances

print("DF of utterances — 1 row per utterance_detail-code")

## [1] "DF of utterances — 1 row per utterance_detail-code"

glimpse(df_insights)

## Rows: 743
## Columns: 13
## $ sid        <fct> 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1…
## $ uid        <fct> 339, 351, 393, 369, 343, 366, 341, 397, 363, 377, 396, 371,…
## $ pid        <fct> j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu2, j2719ee…
## $ TASK       <fct> static, static, static, static, static, static, static, sta…
## $ DATASET    <fct> space, space, space, space, space, space, space, space, spa…
## $ data_order <fct> space-first, space-first, space-first, space-first, space-f…
## $ ixn        <fct> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FAL…
## $ top_code   <fct> DATASET, DATASET, DATASET, VARIABLE, DATASET, VARIABLE, VAR…
## $ mid_code   <fct> NA, NA, NA, distribution (categorical), NA, distribution (c…
## $ low_code   <fct> "data orientation", "data orientation", "data size", "distr…
## $ repns      <chr> "dataframe", "dataframe", "profile", "profile", "profile", …
## $ timestamp  <time> 19:57:00, 20:10:00, 21:27:00, 21:29:00, 21:38:00, 21:44:00…
## $ utterance  <chr> "\"Alright, so every row is the passenger, their home plane…

print("DF of unique utterances — 1 row per utterance [no codes]")

## [1] "DF of unique utterances — 1 row per utterance [no codes]"

glimpse(df_uniques)

## Rows: 662
## Columns: 4
## $ uid     <fct> 339, 351, 393, 369, 343, 366, 341, 397, 363, 377, 396, 371, 40…
## $ pid     <fct> j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu2, j2719eertu…
## $ TASK    <fct> static, static, static, static, static, static, static, static…
## $ DATASET <fct> space, space, space, space, space, space, space, space, space,…

1 DATA PROFILE

TODO TALK WITH DYLAN - resolve missing data in TASK, outcomeType, timestamp… - are these the result of ‘exploded’ utterances that were dual coded? <– need to carry the other attributes across both obs - max of 2 detail-codes applied, correct? - where are the flag codes?

df_insights %>% summarytools::dfSummary(
             plain.ascii  = FALSE,
             graph.magnif = 0.75,
             style        = "grid",
             tmp.img.dir  = "temp",
             missing.col = FALSE, 
             method = "render"
)

1.0.1 Data Frame Summary

1.0.1.1 df_insights

Dimensions: 743 x 13
Duplicates: 0

No	Variable	Stats / Values	Freqs (% of Valid)	Graph	Valid	Missing
1	sid [factor]	1. 0 2. 1 3. 2 4. 3 5. 4 6. 5 7. 6 8. 7 9. 8 10. 9 [ 733 others ]	1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 733 (98.7%)		743 (100.0%)	0 (0.0%)
2	uid [factor]	1. 1 2. 2 3. 3 4. 4 5. 5 6. 6 7. 7 8. 8 9. 9 10. 10 [ 652 others ]	2 ( 0.3%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 1 ( 0.1%) 2 ( 0.3%) 1 ( 0.1%) 1 ( 0.1%) 731 (98.4%)		743 (100.0%)	0 (0.0%)
3	pid [factor]	1. bjs827ee1u 2. 3r2sh20ei 3. 4728sjuiz 4. 7ACC0B75 5. 92ghd48xe 6. iurmer28 7. s294hoei 8. j2719eertu2 9. lkin27js09b 10. li832lin23 [ 3 others ]	29 ( 4.5%) 103 (15.9%) 43 ( 6.6%) 28 ( 4.3%) 56 ( 8.7%) 0 ( 0.0%) 88 (13.6%) 78 (12.1%) 43 ( 6.6%) 51 ( 7.9%) 128 (19.8%)		647 (87.1%)	96 (12.9%)
4	TASK [factor]	1. static 2. ixn	392 (53.4%) 342 (46.6%)		734 (98.8%)	9 (1.2%)
5	DATASET [factor]	1. happiness 2. space	420 (57.2%) 314 (42.8%)		734 (98.8%)	9 (1.2%)
6	data_order [factor]	1. space-first 2. happiness-first 3. NA	300 (40.4%) 434 (58.4%) 9 ( 1.2%)		743 (100.0%)	0 (0.0%)
7	ixn [factor]	1. FALSE 2. TRUE	634 (85.3%) 109 (14.7%)		743 (100.0%)	0 (0.0%)
8	top_code [factor]	1. PROCESS 2. DATASET 3. VARIABLE 4. RELATIONSHIP	160 (21.5%) 176 (23.7%) 122 (16.4%) 285 (38.4%)		743 (100.0%)	0 (0.0%)
9	mid_code [factor]	1. distribution (continuous 2. distribution (categorical 3. relationship (categorical 4. relationship (categorical 5. relationship (continuous 6. relationship (multivariat	77 (18.0%) 54 (12.6%) 28 ( 6.6%) 55 (12.9%) 146 (34.2%) 67 (15.7%)		427 (57.5%)	316 (42.5%)
10	low_code [factor]	1. data orientation 2. data provenance 3. data size 4. distribution outlier (var 5. distribution range [min, 6. distribution shape [shape 7. distribution variance (sd 8. missing data 9. outlier (relationship) 10. plan of action [ 8 others ]	16 ( 2.2%) 11 ( 1.5%) 9 ( 1.2%) 9 ( 1.2%) 33 ( 4.4%) 79 (10.6%) 1 ( 0.1%) 76 (10.2%) 20 ( 2.7%) 52 ( 7.0%) 437 (58.8%)		743 (100.0%)	0 (0.0%)
11	repns [character]	1. scatterplot 2. profile 3. none 4. dataframe 5. Multi-view Chart 6. data_dictionary 7. pairplot 8. lineplot 9. describe 10. double-profiler [ 14 others ]	128 (17.2%) 107 (14.4%) 105 (14.1%) 74 (10.0%) 59 ( 7.9%) 56 ( 7.5%) 50 ( 6.7%) 36 ( 4.8%) 23 ( 3.1%) 23 ( 3.1%) 82 (11.0%)		743 (100.0%)	0 (0.0%)
12	timestamp [hms, difftime]	min : 868 med : 70710 max : 215160 units : secs	622 distinct values		738 (99.3%)	5 (0.7%)
13	utterance [character]	1. [Talking about the profil 2. actually, let me see if p 3. Although we have like les 4. And are they within range 5. And confidence in governm 6. And just I want to see ho 7. And so it looks like it s 8. And then if I had more ti 9. Because it does seem like 10. Data frame. Got a bunch o [ 652 others ]	2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 2 ( 0.3%) 723 (97.3%)		743 (100.0%)	0 (0.0%)

2 UTTERANCES

How many utterances have we collected and coded?

#COUNTS 
n_rows <- df_insights %>% nrow()
n_unique <- nlevels(df_insights$uid)
n_participants <- nlevels(df_insights$pid)

#count number of codes per unique utterance
s <- df_insights %>% group_by(uid) %>% 
    dplyr::summarise(
      count = n()
) %>% arrange(desc(count), .by_group = TRUE)

max_codes <- max(s$count)

#display frequencies
(f <- freq(s$count,
     order    = "freq",
     rows     = 1:10,
     headings = FALSE))

## There are only 2 rows to show; higher numbers will be ignored

## 
##               Freq   % Valid   % Valid Cum.   % Total   % Total Cum.
## ----------- ------ --------- -------------- --------- --------------
##           1    581     87.76          87.76     87.76          87.76
##           2     81     12.24         100.00     12.24         100.00
##        <NA>      0                               0.00         100.00
##       Total    662    100.00         100.00    100.00         100.00

coded_single <- f[1,1]
coded_double <- f[2,1]

There are 743 coded utterances, representing 662 unique statements made by 13 in the study. 581 utterances were single-coded, while 81 utterances received two detail codes. No more than 2 were applied to any single utterance.

2.1 NUMBER OF UTTERANCES

What factors affect how many utterances were produced by participants?

2.1.1 BY Factors

#DEFINE DATAFRAME
df <- df_insights %>% select(pid, uid, TASK, DATASET, data_order, top_code, low_code) 

#SUMMARY TABLE
title = "Utterances by TASK and DATASET"
cols = c("Static Task","Interactive Task","Total Utterances")
cont <- table(df$DATASET, df$TASK)
cont %>% addmargins() %>% kbl(caption = title, col.names = cols) %>%  kable_classic()

Utterances by TASK and DATASET

	Static Task	Interactive Task	Total Utterances
happiness	256	164	420
space	136	178	314
Sum	392	342	734

#MOSAIC PLOTS
# vcd::mosaic(main="Proportion of Utterances by TASK and DATASET",
#             data = df_raw, TASK ~ DATASET, rot_labels=c(0,90,0,0),
#             offset_varnames = c(left = 4.5), offset_labels = c(left = -0.5),just_labels = "right",
#             spacing = spacing_dimequal(unit(1:2, "lines")))
 
# mosaic(formula = ~DATASET + TASK,
#        data = df,
#        main = "Proportion of Utterances by TASK and DATASET",
#        sub = "u = 734 coded utterances",
#        labeling = labeling_values,
#        labeling_args = list(set_varnames = c(graph = "TASK",
#                             datset = "DATASET")))

#DF SUMMARIZED BY TASK + DATASET
df_summary <- df %>% 
  group_by(TASK,DATASET) %>% 
  dplyr::summarise(
    c = n()
  )

## `summarise()` has grouped output by 'TASK'. You can override using the
## `.groups` argument.

#STACKED BAR BY TASK
ggplot(df_summary, aes(x = TASK, y=c, fill= DATASET)) + 
  geom_col() + 
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  labs(title = "Utterances by TASK and DATASET")

#STACKED BAR BY DATASET
ggplot(df_summary, aes(x = DATASET, y=c, fill= TASK)) + 
  geom_col() + 
  scale_fill_brewer(type="qual", palette = 1) +
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") 

2.1.2 By Participant

#DEFINE DATAFRAME
df <- df_insights %>% select(pid, uid, TASK, DATASET, data_order, top_code, low_code) 

#SUMMARY TABLE
title = "Utterances by Participant and TASK"
cols = c("Static Task","Interactive Task","Total Utterances")
cont <- table(df$pid, df$TASK)
cont %>% addmargins() %>% kbl(caption = title, col.names = cols) %>%  kable_classic()

Utterances by Participant and TASK

	Static Task	Interactive Task	Total Utterances
bjs827ee1u	11	18	29
3r2sh20ei	63	40	103
4728sjuiz	30	13	43
7ACC0B75	18	10	28
92ghd48xe	28	28	56
iurmer28	0	0	0
s294hoei	60	28	88
j2719eertu2	32	46	78
lkin27js09b	26	17	43
li832lin23	17	34	51
7382kwtue	24	30	54
E1D39056	10	15	25
8v892iige	27	22	49
Sum	346	301	647

#SUMMARY TABLE
title = "Utterances by Participant and DATASET"
cols = c("Happiness","Space","Total Utterances")
cont <- table(df$pid, df$DATASET)
cont %>% addmargins() %>% kbl(caption = title, col.names = cols) %>%  kable_classic()

Utterances by Participant and DATASET

	Happiness	Space	Total Utterances
bjs827ee1u	11	18	29
3r2sh20ei	63	40	103
4728sjuiz	30	13	43
7ACC0B75	18	10	28
92ghd48xe	28	28	56
iurmer28	0	0	0
s294hoei	60	28	88
j2719eertu2	46	32	78
lkin27js09b	17	26	43
li832lin23	34	17	51
7382kwtue	30	24	54
E1D39056	15	10	25
8v892iige	22	27	49
Sum	374	273	647

#VISUALIZE TASK+DATASET FACET BY PARTICIPANTS
gf_bar( ~ TASK, fill = ~DATASET, data = df) %>% 
gf_facet_grid(~pid)

#UTTERANCES by PARTICPANT and TASK (horizontal)
gf_bar(  pid ~ uid , fill = ~ TASK, data = df) +
# %>%   gf_facet_grid(.~TASK) +
  labs(
    title = "Number of Utterances by Participant and Task",
    subtitle = "some participants were far more talkative than others",
    x = "number of coded utterances",
    y = "participant",
    fill = "Analysis Task"
  )

#UTTERANCES by PARTICPANT and DATASET (horizontal)
gf_bar(  pid ~ uid , fill = ~ DATASET, data = df) + 
  # %>% gf_facet_grid(.~DATASET) +
  labs(
    title = "Number of Utterances by Participant and Dataset",
    subtitle = "Nominal outcome variable (happiness) tended to yield more utterances",
    x = "number of coded utterances",
    y = "participant",
    fill = "Dataset"
  )

#UTTERANCES by PARTICPANT and DATASET (horizontal)
gf_bar(  pid ~ uid , fill = ~ DATASET, data = df) %>% 
  gf_facet_grid(.~TASK) + 
  labs(
    title = "Number of Utterances by Participant, Dataset and Task",
    subtitle = "",
    x = "number of coded utterances",
    y = "participant",
    fill = "DATASET"
  )

2.2 KINDS OF UTTERANCES

2.2.1 BY Factors

#SUMMARIZED DF
df_summary <- df %>% mutate(
  top_code = fct_rev(top_code)) %>%  #reorder
  select( TASK, DATASET, top_code) %>% 
  group_by(TASK,DATASET, top_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#SUMMARY TABLE
title = "CODED Utterances by TYPE and DATASET"
cols = c("Static Task","Interactive Task","Total Utterances")
cont <- table(df$top_code, df$DATASET, df$TASK)
cont %>% addmargins() 

## , ,  = static
## 
##               
##                happiness space Sum
##   PROCESS             70    20  90
##   DATASET             53    44  97
##   VARIABLE            51    25  76
##   RELATIONSHIP        82    47 129
##   Sum                256   136 392
## 
## , ,  = ixn
## 
##               
##                happiness space Sum
##   PROCESS             29    38  67
##   DATASET             27    48  75
##   VARIABLE             8    37  45
##   RELATIONSHIP       100    55 155
##   Sum                164   178 342
## 
## , ,  = Sum
## 
##               
##                happiness space Sum
##   PROCESS             99    58 157
##   DATASET             80    92 172
##   VARIABLE            59    62 121
##   RELATIONSHIP       182   102 284
##   Sum                420   314 734

# %>% kbl(caption = title, col.names = cols) %>%  kable_classic()

#CODED UTTERANCES BAR BY TASK and DATASET
(p_coded_utterances <- 
  ggplot(df_summary, aes(x = TASK, y=c, fill= top_code)) + 
  geom_col() + 
  facet_grid(.~df_summary$DATASET)+
  geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
  scale_fill_brewer(type="qual", palette = 3) +  
  labs(title = "CODED Utterances by TASK and DATASET"))

2.2.2 By Participant

#UTTERANCES by PARTICPANT, TASK, and DATASET (horizontal)
#FACETED BY PARTICIPANT AND TOP CODE 
gf_bar(  TASK ~ uid , fill = ~ DATASET, data = df_insights) %>%  
  gf_facet_grid(top_code ~ pid) + 
  labs(
    title = "High Level Utterances by Participant, Dataset and Dataset",
    subtitle = "",
    x = "number of utterances",
    y = "Analysis Task",
    fill = "Dataset"
  )

## Warning: Ignoring unknown aesthetics: .

#REMOVE NAS
d <- df %>% na.omit()

#TOP CODE by PARTICIPANT & TASK
ggplot (d) +
  geom_bar(aes(y = pid, fill = top_code)) +
  facet_wrap(d$TASK)+
  scale_fill_brewer(type="qual", palette = 3) +
  theme_minimal() + 
  labs(title = "TOP-CODE by TASK")

#TOP CODE by PARTICIPANT & TASK
ggplot (d) +
  geom_bar(aes(y = pid, fill = top_code)) +
  facet_wrap(d$DATASET)+
  scale_fill_brewer(type="qual", palette = 3) +
  theme_minimal() + 
  labs(title = "TOP-CODE by DATASET")

#HACK BIDIRECTIONAL BAR CHART
#SUMMARIZED == reverse static to make bidirectional
# df_summ <- df %>% group_by(pid,TASK,DATASET,top_code,data_order) %>% 
#   dplyr::summarise(
#     n_utterances = n()
#   ) %>% mutate(
#     adj_utterances = ifelse(TASK=="static", (n_utterances*-1), (n_utterances*1))
#   ) %>% arrange(data_order) %>% na.omit()
#   
#  #BIDIRECTIONAL REVERSED
#  ggplot (df_summ, aes(x=adj_utterances, y=pid, fill = top_code)) + 
#   geom_bar(stat="identity") + 
#   facet_wrap(~TASK) +
#   scale_fill_brewer(type="qual", palette = 3) +
#   theme_minimal()
  
# df_hap_first <- df %>% filter(data_order == "happiness-first")
# df_space_first <- df %>% filter(data_order == "space-first")
# 
# top <- gf_bar( pid ~ ., fill = ~top_code, data = df_hap_first) %>%
#   gf_facet_grid(. ~ TASK) +
#   scale_fill_brewer(type="qual", palette = 3)+
#   theme_minimal()
# 
# bottom <- gf_bar( pid ~ ., fill = ~top_code, data = df_space_first) %>%
#   gf_facet_grid(. ~ TASK) +
#   scale_fill_brewer(type="qual", palette = 3)+
#   theme_minimal()
# 
# top
# bottom

df_summary <- df %>% 
group_by(TASK,DATASET, low_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#SUMMARY TABLE
print("DETAIL-CODES TYPE and DATASET") 

## [1] "DETAIL-CODES TYPE and DATASET"

table(df$low_code, df$TASK, df$DATASET) %>% addmargins()

## , ,  = happiness
## 
##                                                     
##                                                      static ixn Sum
##   data orientation                                        2   2   4
##   data provenance                                         3   5   8
##   data size                                               4   1   5
##   distribution outlier (variable)                         4   0   4
##   distribution range [min, max]                          25   3  28
##   distribution shape [shape, skew, kurtosis]             22   4  26
##   distribution variance (sd, var)                         0   1   1
##   missing data                                           23  14  37
##   outlier (relationship)                                  6  12  18
##   plan of action                                         20  10  30
##   relationship cluster(s)/subgroup/ unexpected            7  16  23
##   relationship existence / non-existence                  8   6  14
##   relationship faceted distribution characterization      0   0   0
##   relationship form (linearity/non-linearity)             7   8  15
##   relationship range constriction                         2   3   5
##   relationship strength and/or direction                 52  55 107
##   representation comment                                 50  19  69
##   variable metadata                                      21   5  26
##   Sum                                                   256 164 420
## 
## , ,  = space
## 
##                                                     
##                                                      static ixn Sum
##   data orientation                                        5   5  10
##   data provenance                                         1   2   3
##   data size                                               3   1   4
##   distribution outlier (variable)                         3   2   5
##   distribution range [min, max]                           1   4   5
##   distribution shape [shape, skew, kurtosis]             21  31  52
##   distribution variance (sd, var)                         0   0   0
##   missing data                                           23  14  37
##   outlier (relationship)                                  0   1   1
##   plan of action                                         11   9  20
##   relationship cluster(s)/subgroup/ unexpected            3   3   6
##   relationship existence / non-existence                  8   9  17
##   relationship faceted distribution characterization      0   3   3
##   relationship form (linearity/non-linearity)             0   0   0
##   relationship range constriction                         2   2   4
##   relationship strength and/or direction                 34  37  71
##   representation comment                                  9  29  38
##   variable metadata                                      12  26  38
##   Sum                                                   136 178 314
## 
## , ,  = Sum
## 
##                                                     
##                                                      static ixn Sum
##   data orientation                                        7   7  14
##   data provenance                                         4   7  11
##   data size                                               7   2   9
##   distribution outlier (variable)                         7   2   9
##   distribution range [min, max]                          26   7  33
##   distribution shape [shape, skew, kurtosis]             43  35  78
##   distribution variance (sd, var)                         0   1   1
##   missing data                                           46  28  74
##   outlier (relationship)                                  6  13  19
##   plan of action                                         31  19  50
##   relationship cluster(s)/subgroup/ unexpected           10  19  29
##   relationship existence / non-existence                 16  15  31
##   relationship faceted distribution characterization      0   3   3
##   relationship form (linearity/non-linearity)             7   8  15
##   relationship range constriction                         4   5   9
##   relationship strength and/or direction                 86  92 178
##   representation comment                                 59  48 107
##   variable metadata                                      33  31  64
##   Sum                                                   392 342 734

# #CODED UTTERANCES BAR BY TASK and DATASET
# # (p_coded_utterances <- 
#   ggplot(df_summary, aes(x = TASK, y=c, fill= low_code)) + 
#   geom_col() + 
#   facet_grid(.~df_summary$DATASET)+
#   # geom_text(aes(label=c), size = 3, hjust = 0.5, vjust = 1.5, position = "stack") + 
#   scale_fill_brewer(type="qual", palette = 3) +  
#   labs(title = "CODED Utterances by TASK and DATASET")
#   # )
# 
# 

2.3 PROCESS

df_process <- df %>% filter(top_code == "PROCESS")


summ_process <- df_process %>% 
group_by(TASK,DATASET, low_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#BY TASK
gf_bar(df_process, ~TASK, fill = ~low_code)

#BY PARTICIPANT
gf_bar(df_process, ~TASK, fill = ~low_code) %>% 
  gf_facet_grid(.~pid)

## DATASET

df_dataset <- df %>% filter(top_code == "DATASET")


summ_process <- df_dataset %>% 
group_by(TASK,DATASET, low_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#BY TASK
gf_bar(df_dataset, ~TASK, fill = ~low_code)

#BY PARTICIPANT
gf_bar(df_dataset, ~TASK, fill = ~low_code) %>% 
  gf_facet_grid(.~pid)

## VARIABLE

df_variable <- df %>% filter(top_code == "VARIABLE")


summ_process <- df_variable %>% 
group_by(TASK,DATASET, low_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#BY TASK
gf_bar(df_variable, ~TASK, fill = ~low_code)

#BY PARTICIPANT
gf_bar(df_variable, ~TASK, fill = ~low_code) %>% 
  gf_facet_grid(.~pid)

## RELATIONSHIP

df_relationship <- df %>% filter(top_code == "RELATIONSHIP")


summ_process <- df_relationship %>% 
group_by(TASK,DATASET, low_code) %>% 
  dplyr::summarise( 
    c = n()
  )

## `summarise()` has grouped output by 'TASK', 'DATASET'. You can override using
## the `.groups` argument.

#BY TASK
gf_bar(df_relationship, ~TASK, fill = ~low_code)

#BY PARTICIPANT
gf_bar(df_relationship, ~TASK, fill = ~low_code) %>% 
  gf_facet_grid(.~pid)

# MODELLING

#DEFINE DATAFRAME
df_raw <- df_insights %>% select(pid, uid, TASK, DATASET) %>% mutate(
  TASK = factor(TASK, levels = c("static", "ixn")) #reorder factor levels
) %>% na.omit()
print("WARNING: THE FOLLOWING HAVE OMMITED MISSING DATA RATHER THAN FINDING THE SOURCE")

## [1] "WARNING: THE FOLLOWING HAVE OMMITED MISSING DATA RATHER THAN FINDING THE SOURCE"

#DF SUMMARIZED BY SUBJECT
df_subject <- df_raw %>% group_by(pid, TASK, DATASET) %>% dplyr::summarise(
  n_utterances = n()
)

#VISUALIZE PARTICIPANTS
gf_bar( ~ TASK, fill = ~DATASET, data = df_raw) %>% 
gf_facet_grid(.~pid)

#VISUALIZE TOTALS
gf_bar (~ TASK, fill = ~DATASET, data = df_raw)

gf_bar (~ DATASET, fill = ~TASK, data = df_raw)

#MOSAIC PLOT
vcd::mosaic(main="Proportion of Utterances by TASK and DATASET",
            data = df_raw, TASK ~ DATASET, rot_labels=c(0,90,0,0), 
            offset_varnames = c(left = 4.5), offset_labels = c(left = -0.5),just_labels = "right",
            spacing = spacing_dimequal(unit(1:2, "lines")))

mosaic(formula = ~DATASET + TASK, 
       data = df_raw,
       main = "Proportion of Utterances by TASK and DATASET", 
       sub = "u = 734 utterance codes",
       labeling = labeling_values,
       labeling_args = list(set_varnames = c(graph = "TASK",
                            datset = "DATASET")))

2.4 UTTERANCES

How much variance in number of utterances is explained DATASET, TASK and PARTICIPANT?

2.4.1 OLS Fixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK --> OLS LINEAR REGRESSION
print("OLS-LM, UTTERANCES ~ DATASET + TASK")

## [1] "OLS-LM, UTTERANCES ~ DATASET + TASK"

m1 <- lm(n_utterances ~ DATASET + TASK, data = df_subject)
paste("Model")

## [1] "Model"

summ(m1)

Observations	24
Dependent variable	n_utterances
Type	OLS linear regression

F(2,21)	1.31
R²	0.11
Adj. R²	0.03

	Est.	S.E.	t val.	p
(Intercept)	33.04	4.92	6.71	0.00
DATASETspace	-8.42	5.68	-1.48	0.15
TASKixn	-3.75	5.68	-0.66	0.52
Standard errors: OLS

paste("Partition Variance")

## [1] "Partition Variance"

anova(m1)

## Analysis of Variance Table
## 
## Response: n_utterances
##           Df Sum Sq Mean Sq F value Pr(>F)
## DATASET    1  425.0  425.04  2.1923 0.1536
## TASK       1   84.4   84.38  0.4352 0.5166
## Residuals 21 4071.5  193.88

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(m1)

##                  2.5 %    97.5 %
## (Intercept)   22.80385 43.279487
## DATASETspace -20.23828  3.404950
## TASKixn      -15.57162  8.071617

report(m1) #sanity check

## We fitted a linear model (estimated using OLS) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET + TASK). The model explains a
## statistically not significant and weak proportion of variance (R2 = 0.11, F(2,
## 21) = 1.31, p = 0.290, adj. R2 = 0.03). The model's intercept, corresponding to
## DATASET = happiness and TASK = static, is at 33.04 (95% CI [22.80, 43.28],
## t(21) = 6.71, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -8.42, 95% CI [-20.24, 3.40], t(21) = -1.48, p = 0.154; Std. beta =
## -0.60, 95% CI [-1.43, 0.24])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -3.75, 95% CI [-15.57, 8.07], t(21) = -0.66, p = 0.517; Std. beta = -0.27,
## 95% CI [-1.10, 0.57])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.

plot_model(m1,  show.intercept = TRUE)

check_model(m1)

#NUMBER UTTERANCES predicted by DATASET X TASK --> LINEAR REGRESSION
print("OLS-LM, UTTERANCES ~ DATASET * TASK")

## [1] "OLS-LM, UTTERANCES ~ DATASET * TASK"

m2 <- lm(n_utterances ~ DATASET * TASK, data = df_subject)
paste("Model")

## [1] "Model"

summ(m2)

Observations	24
Dependent variable	n_utterances
Type	OLS linear regression

F(3,20)	1.01
R²	0.13
Adj. R²	0.00

	Est.	S.E.	t val.	p
(Intercept)	35.00	5.76	6.08	0.00
DATASETspace	-12.33	8.14	-1.51	0.15
TASKixn	-7.67	8.14	-0.94	0.36
DATASETspace:TASKixn	7.83	11.52	0.68	0.50
Standard errors: OLS

paste("Partition Variance")

## [1] "Partition Variance"

anova(m2)

## Analysis of Variance Table
## 
## Response: n_utterances
##              Df Sum Sq Mean Sq F value Pr(>F)
## DATASET       1  425.0  425.04  2.1362 0.1594
## TASK          1   84.4   84.38  0.4240 0.5223
## DATASET:TASK  1   92.0   92.04  0.4626 0.5042
## Residuals    20 3979.5  198.98

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(m2)

##                          2.5 %    97.5 %
## (Intercept)           22.98759 47.012415
## DATASETspace         -29.32145  4.654787
## TASKixn              -24.65479  9.321453
## DATASETspace:TASKixn -16.19150 31.858163

report(m2) #sanity check

## We fitted a linear model (estimated using OLS) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET * TASK). The model explains a
## statistically not significant and moderate proportion of variance (R2 = 0.13,
## F(3, 20) = 1.01, p = 0.410, adj. R2 = 9.90e-04). The model's intercept,
## corresponding to DATASET = happiness and TASK = static, is at 35.00 (95% CI
## [22.99, 47.01], t(20) = 6.08, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -12.33, 95% CI [-29.32, 4.65], t(20) = -1.51, p = 0.146; Std. beta =
## -0.87, 95% CI [-2.08, 0.33])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -7.67, 95% CI [-24.65, 9.32], t(20) = -0.94, p = 0.358; Std. beta = -0.54,
## 95% CI [-1.75, 0.66])
##   - The effect of DATASET [space] × TASK [ixn] is statistically non-significant
## and positive (beta = 7.83, 95% CI [-16.19, 31.86], t(20) = 0.68, p = 0.504;
## Std. beta = 0.56, 95% CI [-1.15, 2.26])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.

plot_model(m2,  show.intercept = TRUE)

check_model(m2)

2.4.2 POISSON Fixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK --> POISSON DISTRIBUTION
print("GLM-POISSON, UTTERANCES ~ DATASET + TASK")

## [1] "GLM-POISSON, UTTERANCES ~ DATASET + TASK"

p.1 <- glm(n_utterances ~ DATASET + TASK, data = df_subject, family = "poisson")
paste("Model")

## [1] "Model"

summ(p.1)

Observations	24
Dependent variable	n_utterances
Type	Generalized linear model
Family	poisson
Link	log

𝛘²(2)	18.96
Pseudo-R² (Cragg-Uhler)	0.55
Pseudo-R² (McFadden)	0.07
AIC	264.37
BIC	267.90

	Est.	S.E.	z val.	p
(Intercept)	3.51	0.06	55.32	0.00
DATASETspace	-0.31	0.08	-3.95	0.00
TASKixn	-0.14	0.08	-1.77	0.08
Standard errors: MLE

paste("Partition Variance")

## [1] "Partition Variance"

anova(p.1)

## Analysis of Deviance Table
## 
## Model: poisson, link: log
## 
## Response: n_utterances
## 
## Terms added sequentially (first to last)
## 
## 
##         Df Deviance Resid. Df Resid. Dev
## NULL                       23     156.94
## DATASET  1  15.8313        22     141.11
## TASK     1   3.1324        21     137.98

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(p.1)

## Waiting for profiling to be done...

##                   2.5 %      97.5 %
## (Intercept)   3.3802256  3.62876668
## DATASETspace -0.4715199 -0.15931757
## TASKixn      -0.2941654  0.01495392

report(p.1) #sanity check

## We fitted a poisson model (estimated using ML) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET + TASK). The model's
## explanatory power is substantial (Nagelkerke's R2 = 0.55). The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 3.51
## (95% CI [3.38, 3.63], p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.31, 95% CI [-0.47, -0.16], p < .001; Std. beta = -0.31, 95% CI [-0.47,
## -0.16])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.14, 95% CI [-0.29, 0.01], p = 0.077; Std. beta = -0.14, 95% CI [-0.29,
## 0.01])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.

plot_model(p.1,show.intercept = TRUE)

check_model(p.1)

#NUMBER UTTERANCES predicted by DATASET * TASK --> POISSON DISTRIBUTION
print("GLM-POISSON, UTTERANCES ~ DATASET X TASK")

## [1] "GLM-POISSON, UTTERANCES ~ DATASET X TASK"

p.2 <- glm(n_utterances ~ DATASET * TASK, data = df_subject, family = "poisson")
paste("Model")

## [1] "Model"

summ(p.2)

Observations	24
Dependent variable	n_utterances
Type	Generalized linear model
Family	poisson
Link	log

𝛘²(3)	21.51
Pseudo-R² (Cragg-Uhler)	0.59
Pseudo-R² (McFadden)	0.08
AIC	263.83
BIC	268.54

	Est.	S.E.	z val.	p
(Intercept)	3.56	0.07	51.52	0.00
DATASETspace	-0.43	0.11	-3.95	0.00
TASKixn	-0.25	0.10	-2.37	0.02
DATASETspace:TASKixn	0.25	0.16	1.59	0.11
Standard errors: MLE

paste("Partition Variance")

## [1] "Partition Variance"

anova(p.2)

## Analysis of Deviance Table
## 
## Model: poisson, link: log
## 
## Response: n_utterances
## 
## Terms added sequentially (first to last)
## 
## 
##              Df Deviance Resid. Df Resid. Dev
## NULL                            23     156.94
## DATASET       1  15.8313        22     141.11
## TASK          1   3.1324        21     137.98
## DATASET:TASK  1   2.5434        20     135.44

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(p.2)

## Waiting for profiling to be done...

##                            2.5 %     97.5 %
## (Intercept)           3.41697480  3.6876129
## DATASETspace         -0.65205770 -0.2201745
## TASKixn              -0.45252318 -0.0436738
## DATASETspace:TASKixn -0.05827406  0.5681766

report(p.2) #sanity check

## We fitted a poisson model (estimated using ML) to predict n_utterances with
## DATASET and TASK (formula: n_utterances ~ DATASET * TASK). The model's
## explanatory power is substantial (Nagelkerke's R2 = 0.59). The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 3.56
## (95% CI [3.42, 3.69], p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.43, 95% CI [-0.65, -0.22], p < .001; Std. beta = -0.43, 95% CI [-0.65,
## -0.22])
##   - The effect of TASK [ixn] is statistically significant and negative (beta =
## -0.25, 95% CI [-0.45, -0.04], p = 0.018; Std. beta = -0.25, 95% CI [-0.45,
## -0.04])
##   - The effect of DATASET [space] × TASK [ixn] is statistically non-significant
## and positive (beta = 0.25, 95% CI [-0.06, 0.57], p = 0.111; Std. beta = 0.25,
## 95% CI [-0.06, 0.57])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.

plot_model(p.2,show.intercept = TRUE)

check_model(p.2)

2.4.3 OLS Mixed Effects Models

#NUMBER UTTERANCES predicted by DATASET + TASK | participatnt--> MIXED LINEAR REGRESSION
print("LMER, UTTERANCES ~ DATASET + TASK")

## [1] "LMER, UTTERANCES ~ DATASET + TASK"

mm1 <- lmer(n_utterances ~ DATASET + TASK+ (1|pid), data = df_subject)
paste("Model")

## [1] "Model"

summ(mm1)

Observations	24
Dependent variable	n_utterances
Type	Mixed effects linear regression

AIC	182.77
BIC	188.66
Pseudo-R² (fixed effects)	0.11
Pseudo-R² (total)	0.62

Fixed Effects
	Est.	S.E.	t val.	d.f.	p
(Intercept)	33.04	4.36	7.57	19.70	0.00
DATASETspace	-8.42	3.64	-2.31	10.00	0.04
TASKixn	-3.75	3.64	-1.03	10.00	0.33
p values calculated using Satterthwaite d.f.

Random Effects
Group	Parameter	Std. Dev.
pid	(Intercept)	10.45
Residual		8.91

Grouping Variables
Group	# groups	ICC
pid	12	0.58

paste("Partition Variance")

## [1] "Partition Variance"

anova(mm1)

## Type III Analysis of Variance Table with Satterthwaite's method
##         Sum Sq Mean Sq NumDF DenDF F value  Pr(>F)  
## DATASET 425.04  425.04     1    10  5.3526 0.04324 *
## TASK     84.37   84.37     1    10  1.0625 0.32692  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(mm1)

## Computing profile confidence intervals ...

##                   2.5 %    97.5 %
## .sig01         4.419633 17.646217
## .sigma         5.714441 12.903457
## (Intercept)   24.600922 41.482411
## DATASETspace -15.483056 -1.350277
## TASKixn      -10.816390  3.316389

report(mm1) #sanity check

## We fitted a linear mixed model (estimated using REML and nloptwrap optimizer)
## to predict n_utterances with DATASET and TASK (formula: n_utterances ~ DATASET
## + TASK). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.62) and the
## part related to the fixed effects alone (marginal R2) is of 0.11. The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 33.04
## (95% CI [23.91, 42.17], t(19) = 7.57, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -8.42, 95% CI [-16.03, -0.80], t(19) = -2.31, p = 0.032; Std. beta = -0.60,
## 95% CI [-1.14, -0.06])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -3.75, 95% CI [-11.36, 3.86], t(19) = -1.03, p = 0.316; Std. beta = -0.27,
## 95% CI [-0.81, 0.27])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.

plot_model(mm1,  show.intercept = TRUE)

check_model(mm1)

#NUMBER UTTERANCES predicted by DATASET * TASK  | participatnt--> MIXED LINEAR REGRESSION
print("LMER, UTTERANCES ~ DATASET X TASK")

## [1] "LMER, UTTERANCES ~ DATASET X TASK"

mm2 <- lmer(n_utterances ~ DATASET * TASK + (1|pid), data = df_subject)
paste("Model")

## [1] "Model"

summ(mm2)

Observations	24
Dependent variable	n_utterances
Type	Mixed effects linear regression

AIC	177.31
BIC	184.38
Pseudo-R² (fixed effects)	0.12
Pseudo-R² (total)	0.65

Fixed Effects
	Est.	S.E.	t val.	d.f.	p
(Intercept)	35.00	5.76	6.08	14.69	0.00
DATASETspace	-12.33	8.14	-1.51	14.69	0.15
TASKixn	-7.67	8.14	-0.94	14.69	0.36
DATASETspace:TASKixn	7.83	14.57	0.54	10.00	0.60
p values calculated using Satterthwaite d.f.

Random Effects
Group	Parameter	Std. Dev.
pid	(Intercept)	10.93
Residual		8.91

Grouping Variables
Group	# groups	ICC
pid	12	0.60

paste("Partition Variance")

## [1] "Partition Variance"

anova(mm2)

## Type III Analysis of Variance Table with Satterthwaite's method
##              Sum Sq Mean Sq NumDF DenDF F value  Pr(>F)  
## DATASET      425.04  425.04     1    10  5.3526 0.04324 *
## TASK          84.37   84.37     1    10  1.0625 0.32692  
## DATASET:TASK  22.94   22.94     1    10  0.2889 0.60265  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(mm2)

## Computing profile confidence intervals ...

##                           2.5 %    97.5 %
## .sig01                 4.150279 17.371325
## .sigma                 5.714460 12.902464
## (Intercept)           24.084665 45.915336
## DATASETspace         -27.769948  3.103282
## TASKixn              -23.103281  7.769949
## DATASETspace:TASKixn -20.472642 36.139309

report(mm2) #sanity check

## We fitted a linear mixed model (estimated using REML and nloptwrap optimizer)
## to predict n_utterances with DATASET and TASK (formula: n_utterances ~ DATASET
## * TASK). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.65) and the
## part related to the fixed effects alone (marginal R2) is of 0.12. The model's
## intercept, corresponding to DATASET = happiness and TASK = static, is at 35.00
## (95% CI [22.90, 47.10], t(18) = 6.08, p < .001). Within this model:
## 
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -12.33, 95% CI [-29.44, 4.78], t(18) = -1.51, p = 0.147; Std. beta =
## -0.87, 95% CI [-2.09, 0.34])
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -7.67, 95% CI [-24.78, 9.44], t(18) = -0.94, p = 0.359; Std. beta = -0.54,
## 95% CI [-1.76, 0.67])
##   - The effect of DATASET [space] × TASK [ixn] is statistically non-significant
## and positive (beta = 7.83, 95% CI [-22.78, 38.45], t(18) = 0.54, p = 0.597;
## Std. beta = 0.56, 95% CI [-1.61, 2.72])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald t-distribution approximation.

plot_model(mm2,  show.intercept = TRUE)

check_model(mm2)

2.4.4 POISSON Mixed Effects Models

#NUMBER UTTERANCES predicted by TASK + DATASET  | participatnt--> POISSON MIXED LINEAR REGRESSION
print("POISSON-MER, UTTERANCES ~ DATASET + TASK")

## [1] "POISSON-MER, UTTERANCES ~ DATASET + TASK"

pmm1 <- glmer(n_utterances ~ TASK + DATASET + (1|pid), data = df_subject, family = "poisson")
paste("Model")

## [1] "Model"

summ(pmm1)

Observations	24
Dependent variable	n_utterances
Type	Mixed effects generalized linear model
Family	poisson
Link	log

AIC	189.26
BIC	193.97
Pseudo-R² (fixed effects)	0.12
Pseudo-R² (total)	0.83

Fixed Effects
	Est.	S.E.	z val.	p
(Intercept)	3.41	0.13	25.59	0.00
TASKixn	-0.12	0.08	-1.52	0.13
DATASETspace	-0.31	0.08	-3.87	0.00

Random Effects
Group	Parameter	Std. Dev.
pid	(Intercept)	0.40

Grouping Variables
Group	# groups	ICC
pid	12	0.14

paste("Partition Variance")

## [1] "Partition Variance"

anova(pmm1)

## Analysis of Variance Table
##         npar  Sum Sq Mean Sq F value
## TASK       1  2.9523  2.9523  2.9523
## DATASET    1 14.8233 14.8233 14.8233

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(pmm1)

## Computing profile confidence intervals ...

##                   2.5 %      97.5 %
## .sig01        0.2637590  0.66629252
## (Intercept)   3.1288524  3.68682117
## TASKixn      -0.2776159  0.03519974
## DATASETspace -0.4646941 -0.15157217

report(pmm1) #sanity check

## We fitted a poisson mixed model (estimated using ML and Nelder-Mead optimizer)
## to predict n_utterances with TASK and DATASET (formula: n_utterances ~ TASK +
## DATASET). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.83) and the
## part related to the fixed effects alone (marginal R2) is of 0.12. The model's
## intercept, corresponding to TASK = static and DATASET = happiness, is at 3.41
## (95% CI [3.15, 3.67], p < .001). Within this model:
## 
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.12, 95% CI [-0.28, 0.03], p = 0.128; Std. beta = -0.12, 95% CI [-0.28,
## 0.03])
##   - The effect of DATASET [space] is statistically significant and negative (beta
## = -0.31, 95% CI [-0.46, -0.15], p < .001; Std. beta = -0.31, 95% CI [-0.46,
## -0.15])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.

plot_model(pmm1,  show.intercept = TRUE)

check_model(pmm1)

#NUMBER UTTERANCES predicted by TASK X DATASET  | participatnt--> POISSON MIXED LINEAR REGRESSION
print("POISSON-MER, UTTERANCES ~ DATASET X TASK")

## [1] "POISSON-MER, UTTERANCES ~ DATASET X TASK"

pmm2 <- glmer(n_utterances ~ TASK * DATASET + (1|pid), data = df_subject, family = "poisson")
paste("Model")

## [1] "Model"

summ(pmm2)

Observations	24
Dependent variable	n_utterances
Type	Mixed effects generalized linear model
Family	poisson
Link	log

AIC	191.18
BIC	197.07
Pseudo-R² (fixed effects)	0.13
Pseudo-R² (total)	0.83

Fixed Effects
	Est.	S.E.	z val.	p
(Intercept)	3.45	0.18	19.19	0.00
TASKixn	-0.19	0.26	-0.74	0.46
DATASETspace	-0.38	0.26	-1.46	0.14
TASKixn:DATASETspace	0.14	0.49	0.28	0.78

Random Effects
Group	Parameter	Std. Dev.
pid	(Intercept)	0.40

Grouping Variables
Group	# groups	ICC
pid	12	0.14

paste("Partition Variance")

## [1] "Partition Variance"

anova(pmm2)

## Analysis of Variance Table
##              npar  Sum Sq Mean Sq F value
## TASK            1  3.0017  3.0017  3.0017
## DATASET         1 14.8190 14.8190 14.8190
## TASK:DATASET    1  0.0783  0.0783  0.0783

paste("Confidence Interval on Parameter Estimates")

## [1] "Confidence Interval on Parameter Estimates"

confint(pmm2)

## Computing profile confidence intervals ...

##                           2.5 %    97.5 %
## .sig01                0.2618041 0.6637123
## (Intercept)           3.0601090 3.8198648
## TASKixn              -0.7259280 0.3541708
## DATASETspace         -0.9171433 0.1706822
## TASKixn:DATASETspace -0.9208727 1.1837182

report(pmm2) #sanity check

## We fitted a poisson mixed model (estimated using ML and Nelder-Mead optimizer)
## to predict n_utterances with TASK and DATASET (formula: n_utterances ~ TASK *
## DATASET). The model included pid as random effect (formula: ~1 | pid). The
## model's total explanatory power is substantial (conditional R2 = 0.83) and the
## part related to the fixed effects alone (marginal R2) is of 0.13. The model's
## intercept, corresponding to TASK = static and DATASET = happiness, is at 3.45
## (95% CI [3.10, 3.80], p < .001). Within this model:
## 
##   - The effect of TASK [ixn] is statistically non-significant and negative (beta
## = -0.19, 95% CI [-0.69, 0.31], p = 0.460; Std. beta = -0.19, 95% CI [-0.69,
## 0.31])
##   - The effect of DATASET [space] is statistically non-significant and negative
## (beta = -0.38, 95% CI [-0.88, 0.13], p = 0.145; Std. beta = -0.38, 95% CI
## [-0.88, 0.13])
##   - The effect of TASK [ixn] × DATASET [space] is statistically non-significant
## and positive (beta = 0.14, 95% CI [-0.83, 1.11], p = 0.780; Std. beta = 0.14,
## 95% CI [-0.83, 1.11])
## 
## Standardized parameters were obtained by fitting the model on a standardized
## version of the dataset. 95% Confidence Intervals (CIs) and p-values were
## computed using a Wald z-distribution approximation.

plot_model(pmm2,  show.intercept = TRUE)

check_model(pmm2)

3 TODO REPRESENTATIONS

#COUNTS 
# n_rows <- df_insights %>% nrow()
# n_unique <- nlevels(df_insights$uid)
# n_participants <- nlevels(df_insights$pid)
# 
# #count number of codes per unique utterance
# s <- df_insights %>% group_by(uid) %>% 
#     dplyr::summarise(
#       count = n()
# ) %>% arrange(desc(count), .by_group = TRUE)
# 
# max_codes <- max(s$count)
# 
# #display frequencies
# (f <- freq(s$count,
#      order    = "freq",
#      rows     = 1:10,
#      headings = FALSE))
# 
# coded_single <- f[1,1]
# coded_double <- f[2,1]